Introducing popularity tuning for Similar-Items in Amazon Personalize

Amazon Personalize now allows you to adjust the popularity of your recipe for similar items (aws-similar-items). Similar-Items generates recommendations that are similar to the item a user selects, helping them discover new items in your catalog based on all users’ past behavior and item metadata. Previously, this capability was only available for SIMS, the other Related_Items recipe within Amazon Personalize.

Each customer’s catalog of items and the way users interact with them are unique to their business. When recommending similar items, some customers may want to emphasize popular items more because they increase the likelihood of user interaction, while others may want to de-emphasize popular items to show recommendations that are more similar to the item selected but which are less extensive. known This release gives you more control over the degree to which popularity influences recommendations for similar items, so you can fine-tune the model to meet your particular business needs.

In this post, we show you how to adjust recipe popularity for similar items. We specify a value closer to zero to include more popular articles and a value closer to 1 to place less emphasis on popularity.

Examples of use cases

To explore the impact of this new feature in more detail, let’s review two examples. [1]

First, we used the similar articles recipe to find recommendations similar to the 1994 Disney film The Lion King (IMDB record). When the popularity discount is set to 0, Amazon Personalize recommends movies that have a high frequency (are popular). In this example, the movie Seven (aka Se7en), which occurred 19,295 times in the dataset, is recommended at rank 3.0.

By adjusting the popularity discount to a value of 0.4 for the Lion King recommendations, we see that the Seven movie rank drops to 4.0. We also see films in the Children’s genre such as Babe, Beauty and the Beast, Aladdin and Snow White and the Seven Dwarfs being recommended at a higher rank despite their overall lower popularity in the dataset.

Let’s explore another example. We used the similar articles recipe to find similar recommendations for the 1995 Disney and Pixar film Toy Story ( IMDB record ). When the popularity discount is set to 0, Amazon Personalize recommends movies that have a high frequency in the dataset. In this example, we see that the movie Twelve Monkeys (aka 12 Monkeys), which occurred 6,678 times in the dataset, is recommended at rank 5.0.

By adjusting the popularity discount to a value of 0.4 for Toy Story recommendations, we see that the Twelve Monkeys rank is no longer recommended in the top 10. We also see films in the children’s genre such as Aladdin, Toy Story 2, and A. Bug’s Life is recommended at a higher rank despite its overall lower popularity in the dataset.

Putting more emphasis on the most popular content can help increase the likelihood that users will engage with article recommendations. De-emphasizing popularity can result in recommendations that appear more relevant to the item being searched for, but may be less popular with users. You can adjust the importance given to popularity to meet your business needs for a specific personalization campaign.

Implement popularity adjustment

To adjust the popularity of the recipe for similar items, set the popularity_discount_factor hyperparameter using the AWS Management Console, the AWS SDKs, or the AWS Command Line Interface (AWS CLI).

The following is sample code that sets the popularity discount factor to 0.5 using the AWS SDK:

	response = personalize.create_solution(
		name="movie_lens-with-popularity-discount-0_5".
		recipeARN="arn:aws:personalize:::recipe/aws-similar-items",
		datasetGroupArn=dsg_arn,
		solutionConfig=
			"algorithmHyperParameters" : 
				# set the preferred value of popularity discount here
				"popularity_discount_factor" : "0.50"
			
		
	]

The screenshot below shows setting the popularity discount factor to 0.3 in the Amazon Personalize console.

conclusion

With the popularity adjustment, you can now further refine the recipe for similar items in Amazon Personalize to control the degree to which popularity influences item recommendations. This gives you more control over defining the end-user experience and what is included or excluded from your similar article recommendations.

For more details on how to implement popularity adjustment for the recipe of similar items, see the documentation.

References

About the Authors

Julia McCombs Clark is a CTO on the Amazon Personalize team.

Nihal Harish is a software development engineer on the Amazon Personalize team.